#!/usr/bin/python3 """Test suite for the method to extract text from an html string""" from ..html_tools import html_to_text def test_html_to_text_func(): test_html = """ Some initial text

Which is across multiple lines

More Text
So let's see what happens.
Even More Text """ # extract text, with 'render_anchor_tag_content' set to False text_content = html_to_text(test_html, render_anchor_tag_content=False) no_links_text = \ "Some initial text\n\nWhich is across multiple " \ "lines\n\nMore Text So let's see what happens. Even More Text" # check that no links are in the extracted text assert text_content == no_links_text # extract text, with 'render_anchor_tag_content' set to True text_content = html_to_text(test_html, render_anchor_tag_content=True) links_text = \ "Some initial text\n\nWhich is across multiple lines\n\n[ More Text " \ "](/first_link) So let's see what happens. [ Even More Text ]" \ "(second_link.com)" # check that links are present in the extracted text assert text_content == links_text